Last week my post was on the definition of a recession and argued against using the “two quarters of declining GDP standard.” Little did I know that the very next day, the White House’s Council of Economic Advisors would write a blog post on this topic the very next day (essentially taking the same position as I did). The CEA post set of a long discussion on Twitter, which even spilled over into the national media.
I don’t want to get into that debate here today. Instead, let’s look at the history of dating business cycles, specifically in the 19th century. Forget waiting a few months or even a year for an official NBER announcement: the first attempt to date business cycles was going back over 100 years! In going over this history, perhaps we can learn something about our current debates over recessions, but I think the history is interesting in its own right (it’s also a great example of how we can get better data and use it to answer important questions).
I’ll give a brief history here, but read this Romer and Romer conference paper to get an excellent, full history of the NBER’s business cycle dating. The NBER was essentially found as an institution to study business cycles. One of the first major publications was Willard Thorp’s Business Annals, published in 1926. It was groundbreaking study, which not only provided annual business cycle dates for the entire history of the US, it also did so for 16 other countries for roughly the same time period!
While such an undertaking was impressive, the methods used were pretty unsophisticated from the hindsight of almost 100 years later. First, these are annual estimates, not monthly or even quarterly. Monthly estimates would come later, first appearing in Burns and Mitchell’s 1946 volume Measuring Business Cycles. Those monthly estimates began in 1854, and there are the same ones you will find on the NBER website today, essentially unmodified by even a single month for the late 19th century.
But what of the first half of the 19th century? How did Thorp date recessions?
Let’s look at a specific time period to get a sense of Thorp’s analysis. Thorp lists there being 5 separate recessions between 1836 and 1855, with one lasting for 4 years (1839-1843)! The economy is apparently in recession for parts of at half of the time for these 20 years. And this is an extremely important era to study, following the closing of the Second Bank of the United States in 1836.
Here’s what Thorp did: he read old newspapers, looked at the history of the year, and then… essentially made a judgement call. A few examples from the late 1830s:
Now, he’s certainly looking at data here when he can. Prices, production, specie payments, foreign trade. It’s all in there. But really, he’s just eyeballing it. A groundbreaking analysis, but surely there must be a more consistent way to do this?
It wasn’t until almost 80 years later that we would get a major improvement on these dates (the 2006 edition of the Historical Statistics of the United States still uses Thorp’s dates for the early 19th century). In his dissertation and then a series of academic papers, Joseph H. Davis created an actual index of industrial production for the US before 1915, essentially attempting to duplicate the modern industrial production index from the Federal Reserve, which starts in 1919.
Davis published the index in the Quarterly Journal of Economics in 2004 and also published a revised business cycle chronology in the Journal of Economic History in 2006 (two very nice publications to come out of your dissertation!). What does he find? Let’s first look at the index for the 1830s and 1840s, as seen at the right (from Table 1). It’s a true index, with a single number for each year measuring the industrial output of the economy. If it goes up from year to year, we’ll call that an economic expansion. If it goes down, that’s a recession. Easy enough!
Note also: there were already existing estimates of GDP for this same time period, but these estimates were never intended to measure year-to-year fluctuations in the economy (even though they did decline some years). These GDP estimates were designed to look at long-run growth trends.
How does Davis construct his index? The details are in his QJE article if you want to read all of them, but I want emphasize one important point, especially for young researchers: all of the data that Davis used in the 2000s was available in the 1920s! Well, sort of. Some of it has been better assembled today than it was in the 1920s. Davis uses a mix of government records of output in various industries, as well as private records of production. One example of a private record is a database of 120,000 railroad engines, which he was able to cross-reference with thousands of manufacturers of railroad companies to see when they were built. These are the kinds of sources and techniques Davis uses to construct his index. But much of that data had not been put in one place when Thorp performed his analysis in the 1920s, so Davis clearly had an advantage (and he used it!).
So what’s the payoff to all this new data? The big result, which Davis emphasizes in his JEH article, is that the 19th century was much less volatile, and less prone to recessions, than the Thorp data indicates. Many recessions are shortened or removed altogether. For example, in that 1836-1855 period, 3 of the 5 recessions are eliminated entirely, since the industrial production index increased in those year. And the other two recessions are shortened to just one year each. There are similar changes throughout the 19th century and early 20th century. 21 of the 29 business cycles in the original Thorp chronology are altered in some way, with 8 being eliminated altogether, and the remainder are shortened in some length (none are longer in the Davis chronology). The famous “Long Depression” of 1873 to 1878 is shortened by 3 years by Davis (he says it ends in 1875).
Overall, these changes suggest that not only is the 19th century much less volatile than we thought, it compares well with the late 20th century, the period of intensive business cycle management through both monetary and fiscal policy (see this paper by Selgin, Lastrapes, and White which expands on this point considering the role of the Fed). Comparing Davis’ IP Index in the 19th century to the Fed’s IP Index in the late 20th century, business cycles are of similar frequency and duration (Table 4 in the paper). In the phrase of Christina Romer, there was “spurious volatility” in the 19th century business cycle data.
Like Thorp’s work in the 1920s, Davis’ work was indeed groundbreaking, and should fundamentally change how we think about economic fluctuations and recessions in the 19th century. The Davis chronlogy is different from how NBER dates business cycles today, in that it relies on just one measure (industrial production) rather than the suite of measures NBER uses today. But industrial production is still one of those measures NBER’s Business Cycle Dating Committee uses! Also, like Thorp, Davis is only providing annual dates for recessions, rather than the monthly dates we have today.
But even with these differences, the work by Davis should remind us that our measurement of business cycles is subject to constant improvement, not only as new and updated data become available, but also as we have better ways of thinking about the economy. Knowing in real time how the economy is doing is important for many reasons (for both the public and private sectors), but the dates in the NBER series aren’t exactly designed to provide a real-time judgment on the economy.